1 /*- 2 * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU) 3 * 4 * Copyright (c) 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * The Mach Operating System project at Carnegie-Mellon University. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * from: @(#)vm_glue.c 8.6 (Berkeley) 1/5/94 35 * 36 * 37 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 38 * All rights reserved. 39 * 40 * Permission to use, copy, modify and distribute this software and 41 * its documentation is hereby granted, provided that both the copyright 42 * notice and this permission notice appear in all copies of the 43 * software, derivative works or modified versions, and any portions 44 * thereof, and that both notices appear in supporting documentation. 45 * 46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 49 * 50 * Carnegie Mellon requests users of this software to return to 51 * 52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 53 * School of Computer Science 54 * Carnegie Mellon University 55 * Pittsburgh PA 15213-3890 56 * 57 * any improvements or extensions that they make and grant Carnegie the 58 * rights to redistribute these changes. 59 */ 60 61 #include <sys/cdefs.h> 62 #include "opt_vm.h" 63 #include "opt_kstack_pages.h" 64 #include "opt_kstack_max_pages.h" 65 #include "opt_kstack_usage_prof.h" 66 67 #include <sys/param.h> 68 #include <sys/systm.h> 69 #include <sys/asan.h> 70 #include <sys/domainset.h> 71 #include <sys/limits.h> 72 #include <sys/lock.h> 73 #include <sys/malloc.h> 74 #include <sys/msan.h> 75 #include <sys/mutex.h> 76 #include <sys/proc.h> 77 #include <sys/racct.h> 78 #include <sys/refcount.h> 79 #include <sys/resourcevar.h> 80 #include <sys/rwlock.h> 81 #include <sys/sched.h> 82 #include <sys/sf_buf.h> 83 #include <sys/shm.h> 84 #include <sys/smp.h> 85 #include <sys/vmmeter.h> 86 #include <sys/vmem.h> 87 #include <sys/sx.h> 88 #include <sys/sysctl.h> 89 #include <sys/kernel.h> 90 #include <sys/ktr.h> 91 #include <sys/unistd.h> 92 93 #include <vm/uma.h> 94 #include <vm/vm.h> 95 #include <vm/vm_param.h> 96 #include <vm/pmap.h> 97 #include <vm/vm_domainset.h> 98 #include <vm/vm_map.h> 99 #include <vm/vm_page.h> 100 #include <vm/vm_pageout.h> 101 #include <vm/vm_object.h> 102 #include <vm/vm_kern.h> 103 #include <vm/vm_extern.h> 104 #include <vm/vm_pager.h> 105 #include <vm/swap_pager.h> 106 107 #include <machine/cpu.h> 108 109 /* 110 * MPSAFE 111 * 112 * WARNING! This code calls vm_map_check_protection() which only checks 113 * the associated vm_map_entry range. It does not determine whether the 114 * contents of the memory is actually readable or writable. In most cases 115 * just checking the vm_map_entry is sufficient within the kernel's address 116 * space. 117 */ 118 int 119 kernacc(void *addr, int len, int rw) 120 { 121 boolean_t rv; 122 vm_offset_t saddr, eaddr; 123 vm_prot_t prot; 124 125 KASSERT((rw & ~VM_PROT_ALL) == 0, 126 ("illegal ``rw'' argument to kernacc (%x)\n", rw)); 127 128 if ((vm_offset_t)addr + len > vm_map_max(kernel_map) || 129 (vm_offset_t)addr + len < (vm_offset_t)addr) 130 return (FALSE); 131 132 prot = rw; 133 saddr = trunc_page((vm_offset_t)addr); 134 eaddr = round_page((vm_offset_t)addr + len); 135 vm_map_lock_read(kernel_map); 136 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); 137 vm_map_unlock_read(kernel_map); 138 return (rv == TRUE); 139 } 140 141 /* 142 * MPSAFE 143 * 144 * WARNING! This code calls vm_map_check_protection() which only checks 145 * the associated vm_map_entry range. It does not determine whether the 146 * contents of the memory is actually readable or writable. vmapbuf(), 147 * vm_fault_quick(), or copyin()/copout()/su*()/fu*() functions should be 148 * used in conjunction with this call. 149 */ 150 int 151 useracc(void *addr, int len, int rw) 152 { 153 boolean_t rv; 154 vm_prot_t prot; 155 vm_map_t map; 156 157 KASSERT((rw & ~VM_PROT_ALL) == 0, 158 ("illegal ``rw'' argument to useracc (%x)\n", rw)); 159 prot = rw; 160 map = &curproc->p_vmspace->vm_map; 161 if ((vm_offset_t)addr + len > vm_map_max(map) || 162 (vm_offset_t)addr + len < (vm_offset_t)addr) { 163 return (FALSE); 164 } 165 vm_map_lock_read(map); 166 rv = vm_map_check_protection(map, trunc_page((vm_offset_t)addr), 167 round_page((vm_offset_t)addr + len), prot); 168 vm_map_unlock_read(map); 169 return (rv == TRUE); 170 } 171 172 int 173 vslock(void *addr, size_t len) 174 { 175 vm_offset_t end, last, start; 176 vm_size_t npages; 177 int error; 178 179 last = (vm_offset_t)addr + len; 180 start = trunc_page((vm_offset_t)addr); 181 end = round_page(last); 182 if (last < (vm_offset_t)addr || end < (vm_offset_t)addr) 183 return (EINVAL); 184 npages = atop(end - start); 185 if (npages > vm_page_max_user_wired) 186 return (ENOMEM); 187 error = vm_map_wire(&curproc->p_vmspace->vm_map, start, end, 188 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 189 if (error == KERN_SUCCESS) { 190 curthread->td_vslock_sz += len; 191 return (0); 192 } 193 194 /* 195 * Return EFAULT on error to match copy{in,out}() behaviour 196 * rather than returning ENOMEM like mlock() would. 197 */ 198 return (EFAULT); 199 } 200 201 void 202 vsunlock(void *addr, size_t len) 203 { 204 205 /* Rely on the parameter sanity checks performed by vslock(). */ 206 MPASS(curthread->td_vslock_sz >= len); 207 curthread->td_vslock_sz -= len; 208 (void)vm_map_unwire(&curproc->p_vmspace->vm_map, 209 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), 210 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 211 } 212 213 /* 214 * Pin the page contained within the given object at the given offset. If the 215 * page is not resident, allocate and load it using the given object's pager. 216 * Return the pinned page if successful; otherwise, return NULL. 217 */ 218 static vm_page_t 219 vm_imgact_hold_page(vm_object_t object, vm_ooffset_t offset) 220 { 221 vm_page_t m; 222 vm_pindex_t pindex; 223 224 pindex = OFF_TO_IDX(offset); 225 (void)vm_page_grab_valid_unlocked(&m, object, pindex, 226 VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED); 227 return (m); 228 } 229 230 /* 231 * Return a CPU private mapping to the page at the given offset within the 232 * given object. The page is pinned before it is mapped. 233 */ 234 struct sf_buf * 235 vm_imgact_map_page(vm_object_t object, vm_ooffset_t offset) 236 { 237 vm_page_t m; 238 239 m = vm_imgact_hold_page(object, offset); 240 if (m == NULL) 241 return (NULL); 242 sched_pin(); 243 return (sf_buf_alloc(m, SFB_CPUPRIVATE)); 244 } 245 246 /* 247 * Destroy the given CPU private mapping and unpin the page that it mapped. 248 */ 249 void 250 vm_imgact_unmap_page(struct sf_buf *sf) 251 { 252 vm_page_t m; 253 254 m = sf_buf_page(sf); 255 sf_buf_free(sf); 256 sched_unpin(); 257 vm_page_unwire(m, PQ_ACTIVE); 258 } 259 260 void 261 vm_sync_icache(vm_map_t map, vm_offset_t va, vm_offset_t sz) 262 { 263 264 pmap_sync_icache(map->pmap, va, sz); 265 } 266 267 vm_object_t kstack_object; 268 static uma_zone_t kstack_cache; 269 static int kstack_cache_size; 270 271 static int 272 sysctl_kstack_cache_size(SYSCTL_HANDLER_ARGS) 273 { 274 int error, oldsize; 275 276 oldsize = kstack_cache_size; 277 error = sysctl_handle_int(oidp, arg1, arg2, req); 278 if (error == 0 && req->newptr && oldsize != kstack_cache_size) 279 uma_zone_set_maxcache(kstack_cache, kstack_cache_size); 280 return (error); 281 } 282 SYSCTL_PROC(_vm, OID_AUTO, kstack_cache_size, 283 CTLTYPE_INT|CTLFLAG_MPSAFE|CTLFLAG_RW, &kstack_cache_size, 0, 284 sysctl_kstack_cache_size, "IU", "Maximum number of cached kernel stacks"); 285 286 /* 287 * Create the kernel stack (including pcb for i386) for a new thread. 288 */ 289 static vm_offset_t 290 vm_thread_stack_create(struct domainset *ds, int pages) 291 { 292 vm_page_t ma[KSTACK_MAX_PAGES]; 293 vm_offset_t ks; 294 int i; 295 296 /* 297 * Get a kernel virtual address for this thread's kstack. 298 */ 299 ks = kva_alloc((pages + KSTACK_GUARD_PAGES) * PAGE_SIZE); 300 if (ks == 0) { 301 printf("%s: kstack allocation failed\n", __func__); 302 return (0); 303 } 304 305 if (KSTACK_GUARD_PAGES != 0) { 306 pmap_qremove(ks, KSTACK_GUARD_PAGES); 307 ks += KSTACK_GUARD_PAGES * PAGE_SIZE; 308 } 309 310 /* 311 * Allocate physical pages to back the stack. 312 */ 313 vm_thread_stack_back(ds, ks, ma, pages, VM_ALLOC_NORMAL); 314 for (i = 0; i < pages; i++) 315 vm_page_valid(ma[i]); 316 pmap_qenter(ks, ma, pages); 317 318 return (ks); 319 } 320 321 static void 322 vm_thread_stack_dispose(vm_offset_t ks, int pages) 323 { 324 vm_page_t m; 325 vm_pindex_t pindex; 326 int i; 327 328 pindex = atop(ks - VM_MIN_KERNEL_ADDRESS); 329 330 pmap_qremove(ks, pages); 331 VM_OBJECT_WLOCK(kstack_object); 332 for (i = 0; i < pages; i++) { 333 m = vm_page_lookup(kstack_object, pindex + i); 334 if (m == NULL) 335 panic("%s: kstack already missing?", __func__); 336 vm_page_xbusy_claim(m); 337 vm_page_unwire_noq(m); 338 vm_page_free(m); 339 } 340 VM_OBJECT_WUNLOCK(kstack_object); 341 kasan_mark((void *)ks, ptoa(pages), ptoa(pages), 0); 342 kva_free(ks - (KSTACK_GUARD_PAGES * PAGE_SIZE), 343 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE); 344 } 345 346 /* 347 * Allocate the kernel stack for a new thread. 348 */ 349 int 350 vm_thread_new(struct thread *td, int pages) 351 { 352 vm_offset_t ks; 353 354 /* Bounds check */ 355 if (pages <= 1) 356 pages = kstack_pages; 357 else if (pages > KSTACK_MAX_PAGES) 358 pages = KSTACK_MAX_PAGES; 359 360 ks = 0; 361 if (pages == kstack_pages && kstack_cache != NULL) 362 ks = (vm_offset_t)uma_zalloc(kstack_cache, M_NOWAIT); 363 364 /* 365 * Ensure that kstack objects can draw pages from any memory 366 * domain. Otherwise a local memory shortage can block a process 367 * swap-in. 368 */ 369 if (ks == 0) 370 ks = vm_thread_stack_create(DOMAINSET_PREF(PCPU_GET(domain)), 371 pages); 372 if (ks == 0) 373 return (0); 374 td->td_kstack = ks; 375 td->td_kstack_pages = pages; 376 kasan_mark((void *)ks, ptoa(pages), ptoa(pages), 0); 377 kmsan_mark((void *)ks, ptoa(pages), KMSAN_STATE_UNINIT); 378 return (1); 379 } 380 381 /* 382 * Dispose of a thread's kernel stack. 383 */ 384 void 385 vm_thread_dispose(struct thread *td) 386 { 387 vm_offset_t ks; 388 int pages; 389 390 pages = td->td_kstack_pages; 391 ks = td->td_kstack; 392 td->td_kstack = 0; 393 td->td_kstack_pages = 0; 394 kasan_mark((void *)ks, 0, ptoa(pages), KASAN_KSTACK_FREED); 395 if (pages == kstack_pages) 396 uma_zfree(kstack_cache, (void *)ks); 397 else 398 vm_thread_stack_dispose(ks, pages); 399 } 400 401 /* 402 * Allocate physical pages, following the specified NUMA policy, to back a 403 * kernel stack. 404 */ 405 void 406 vm_thread_stack_back(struct domainset *ds, vm_offset_t ks, vm_page_t ma[], 407 int npages, int req_class) 408 { 409 vm_pindex_t pindex; 410 int n; 411 412 pindex = atop(ks - VM_MIN_KERNEL_ADDRESS); 413 414 VM_OBJECT_WLOCK(kstack_object); 415 for (n = 0; n < npages;) { 416 if (vm_ndomains > 1) 417 kstack_object->domain.dr_policy = ds; 418 419 /* 420 * Use WAITFAIL to force a reset of the domain selection policy 421 * if we had to sleep for pages. 422 */ 423 n += vm_page_grab_pages(kstack_object, pindex + n, 424 req_class | VM_ALLOC_WIRED | VM_ALLOC_WAITFAIL, 425 &ma[n], npages - n); 426 } 427 VM_OBJECT_WUNLOCK(kstack_object); 428 } 429 430 static int 431 kstack_import(void *arg, void **store, int cnt, int domain, int flags) 432 { 433 struct domainset *ds; 434 int i; 435 436 if (domain == UMA_ANYDOMAIN) 437 ds = DOMAINSET_RR(); 438 else 439 ds = DOMAINSET_PREF(domain); 440 441 for (i = 0; i < cnt; i++) { 442 store[i] = (void *)vm_thread_stack_create(ds, kstack_pages); 443 if (store[i] == NULL) 444 break; 445 } 446 return (i); 447 } 448 449 static void 450 kstack_release(void *arg, void **store, int cnt) 451 { 452 vm_offset_t ks; 453 int i; 454 455 for (i = 0; i < cnt; i++) { 456 ks = (vm_offset_t)store[i]; 457 vm_thread_stack_dispose(ks, kstack_pages); 458 } 459 } 460 461 static void 462 kstack_cache_init(void *null) 463 { 464 kstack_object = vm_object_allocate(OBJT_SWAP, 465 atop(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS)); 466 kstack_cache = uma_zcache_create("kstack_cache", 467 kstack_pages * PAGE_SIZE, NULL, NULL, NULL, NULL, 468 kstack_import, kstack_release, NULL, 469 UMA_ZONE_FIRSTTOUCH); 470 kstack_cache_size = imax(128, mp_ncpus * 4); 471 uma_zone_set_maxcache(kstack_cache, kstack_cache_size); 472 } 473 SYSINIT(vm_kstacks, SI_SUB_KMEM, SI_ORDER_ANY, kstack_cache_init, NULL); 474 475 #ifdef KSTACK_USAGE_PROF 476 /* 477 * Track maximum stack used by a thread in kernel. 478 */ 479 static int max_kstack_used; 480 481 SYSCTL_INT(_debug, OID_AUTO, max_kstack_used, CTLFLAG_RD, 482 &max_kstack_used, 0, 483 "Maximum stack depth used by a thread in kernel"); 484 485 void 486 intr_prof_stack_use(struct thread *td, struct trapframe *frame) 487 { 488 vm_offset_t stack_top; 489 vm_offset_t current; 490 int used, prev_used; 491 492 /* 493 * Testing for interrupted kernel mode isn't strictly 494 * needed. It optimizes the execution, since interrupts from 495 * usermode will have only the trap frame on the stack. 496 */ 497 if (TRAPF_USERMODE(frame)) 498 return; 499 500 stack_top = td->td_kstack + td->td_kstack_pages * PAGE_SIZE; 501 current = (vm_offset_t)(uintptr_t)&stack_top; 502 503 /* 504 * Try to detect if interrupt is using kernel thread stack. 505 * Hardware could use a dedicated stack for interrupt handling. 506 */ 507 if (stack_top <= current || current < td->td_kstack) 508 return; 509 510 used = stack_top - current; 511 for (;;) { 512 prev_used = max_kstack_used; 513 if (prev_used >= used) 514 break; 515 if (atomic_cmpset_int(&max_kstack_used, prev_used, used)) 516 break; 517 } 518 } 519 #endif /* KSTACK_USAGE_PROF */ 520 521 /* 522 * Implement fork's actions on an address space. 523 * Here we arrange for the address space to be copied or referenced, 524 * allocate a user struct (pcb and kernel stack), then call the 525 * machine-dependent layer to fill those in and make the new process 526 * ready to run. The new process is set up so that it returns directly 527 * to user mode to avoid stack copying and relocation problems. 528 */ 529 int 530 vm_forkproc(struct thread *td, struct proc *p2, struct thread *td2, 531 struct vmspace *vm2, int flags) 532 { 533 struct proc *p1 = td->td_proc; 534 struct domainset *dset; 535 int error; 536 537 if ((flags & RFPROC) == 0) { 538 /* 539 * Divorce the memory, if it is shared, essentially 540 * this changes shared memory amongst threads, into 541 * COW locally. 542 */ 543 if ((flags & RFMEM) == 0) { 544 error = vmspace_unshare(p1); 545 if (error) 546 return (error); 547 } 548 cpu_fork(td, p2, td2, flags); 549 return (0); 550 } 551 552 if (flags & RFMEM) { 553 p2->p_vmspace = p1->p_vmspace; 554 refcount_acquire(&p1->p_vmspace->vm_refcnt); 555 } 556 dset = td2->td_domain.dr_policy; 557 while (vm_page_count_severe_set(&dset->ds_mask)) { 558 vm_wait_doms(&dset->ds_mask, 0); 559 } 560 561 if ((flags & RFMEM) == 0) { 562 p2->p_vmspace = vm2; 563 if (p1->p_vmspace->vm_shm) 564 shmfork(p1, p2); 565 } 566 567 /* 568 * cpu_fork will copy and update the pcb, set up the kernel stack, 569 * and make the child ready to run. 570 */ 571 cpu_fork(td, p2, td2, flags); 572 return (0); 573 } 574 575 /* 576 * Called after process has been wait(2)'ed upon and is being reaped. 577 * The idea is to reclaim resources that we could not reclaim while 578 * the process was still executing. 579 */ 580 void 581 vm_waitproc(struct proc *p) 582 { 583 584 vmspace_exitfree(p); /* and clean-out the vmspace */ 585 } 586 587 void 588 kick_proc0(void) 589 { 590 591 wakeup(&proc0); 592 } 593